I recall talking with an EMC Sales Engineer awhile ago about how HDS was the only company out there with a similar model to EMC in storage but they were still several
generations behind in software (ie. I believe EMC was then on its 4th Gen SRDF, while HDS still only had 1st level functionality).
The original SRDF that EMC introduced in 1995 broke new ground because it utilized array-based processing power but it was limited to a source-target (one-way) model and could only support expensive leased lines to establish synchronous (real-time) remote mirrors.
Hitachi's mainframe and mainframe storage design philosophy was a replica of IBM's design philosophy until IBM swallowed its considerable pride and entered into a deal with a fierce piece-part rival,StorageTek, to resell IceBerg (log-structured system with compression) in 1997. Then it shifted to emulating EMC, typically dangling the prospects of a HMTT (now bankrupt) type of IPO for HDS and pointing to minor technical advances as proof of its technical lead. Hitachi’s product is patterned after SRDF in that it also uses array-based intelligence and it is more evolved than the original SRDF but not by much. HDS only released its SRDF-clone last year and had to subcontract their Celerra-clone only last June so HDS is behind by more than 4 years.
As an aside, IBM’s historical problem with EMC's approach is that array-based intelligence siphons valuable mainframe/high-end Unix MIPS away from its mainframes and servers. EMC, for example, has traditionally complemented its array-based intelligence with management consoles that run on commodity NT or Unix servers or laptops and are tightly integerated with EMC's call-home diagnostics technology -- SRDF server, TimeFinder server, EDM server, FastTrak server, ControlCenter cluster servers, etc.
Understanding the way IBM pulls in revenue with every mainframe is useful. Right now, IBM is benefiting from a new mainframe upgrade cycle whose scale should become clearer at year's end. The barebones mainframe market shrunk from around $14B in 1990 to around $1B in 2000. Mainframe storage shrunk from $8B in 1990 to $1.9B in 2000. Despite the widespread migration away from mainframes, each mainframe MIPS dollar still pulls in more than $7 in mainframe sofware sales and an undisclosed amount of planning/break/fix service sales. Mainframes are engineered to last 7 years and are typically depreciated over 5 years so the service contract may be the most lucrative part of the revenue stream over the life of each mainframe.
Until it changed its software pricing policy late last year to the same per-capacity policy used by Sun, Compaq and HWP, EMC priced its software on a per-Symm basis. IBM was frequently caught off-balance in the marketplace because it could not match EMC's technical advances and it could not afford to cannibalize its mainframe-related revenues to compete with EMC's value proposition, particularly EMC's software which was priced on a per-Symm basis. This is the same problem faced by the leading high-end Unix server vendors like Sun and HWP. That is why EMC routinely controls more than 40% of the S390, Solaris and HP-UX market and why price is typically the main weapon used to stem EMC's advances in those markets.
Going back, the current version of SRDF is bi-directional; i.e., the source can also be the target. It also supports all types of networks to establish synchronous (real-time) and asynchronous (variable time) remote mirrors at the block and file levels for the Symmetrix and Celerra platforms.
This means that EMC can use one Symm -- stand-alone or as part of a SAN -- to create multiple synchronous and asynchronous copies in different Symms -- again, stand-alone or as part of a SAN -- in different locations as well as receive multiple synchronous and asynchronous copies from different Symms -- stand-alone or as part of a SAN -- in different locations using the most cost-effective combination of leased lines (guaranteed performance) and IP networks (best efforts performance) to address business requirements like network caching.
EMC has thoroughly integrated SRDF with its very popular TimeFinder, which creates volume mirrors inside each Symm for parallel processing. Before EMC changed its software pricing policy last year, it had already sold more than 9,000 SRDF licenses and more than 15,000 TimeFinder licenses. SRDF came out in 1995 and TimeFinder came out in 1998.
The main criticism of SRDF typically comes from software- only vendors and centers around the requirement of a Symm at both sides of each real-time or variable-time mirror.
But this ignores the growing importance of system level software that becomes easier to implement as embedded chip and programmable chip technology become cheaper, faster and better with every turn of Moore's Law.
.....Reliability and robustness against data errors is a requirement for storage systems. System interconnects, such as Ethernet and InfiniBand, typically use parity or CRC mechanisms to detect failed bits, usually resorting to software mechanisms for error-recovery. Software-recovery places a complexity burden on system designers and requires extra latency. This is usually not a problem for system-to- system interfaces, but within the system -- where real-time response is critical -- software error-recovery is unacceptable....
is.pennnet.com
This also ignores the fact that nobody would go through the trouble of investing in disaster recovery if they use different storage systems with different failure management schemes inside each box. In fact, RAS (reliability, availability and scalability) features are the most resistant to standardization because each vendor has a different approach to system design and makes different system design trade-offs for different market segments.
For example, each Symmetrix has a dedicated service processor that runs more than 1,000 diagnostic routines every 2 hours with remote monitoring and same day parts replacement services provided by EMC Global Services. In a simple SRDF installation in a raised floor environment, that means the both sides of the real-time mirror are being monitored by the same vendor under the same service contract. Hitachi and IBM have different approaches to remote monitoring that are not yet as sophisticated as EMC in storage (electronic and mechanical failure profiles) because they kept on changing monolithic storage platforms over the last 10 years. Now, why would anyone add several layers of complexity by putting a Symm and a Shark or Lightning at opposite sides of a real-time mirror. What business problem does that piecemeal approach solve?
The fast-growing replication market is going EMC's way. EMC’s replication revenue consistently dwarfs the replication revenues of its next 3 or 4 competitors combined! Because IBM's disaster recovery product is tightly integrated into its mainframes, IBM is its own biggest customer for its disaster recovery products. IBM, Sunguard and Comdisco control more than 80% of the disaster recovery services market. |
